It goes without saying that a meaningful study should have a random
selection process, although it happens all the time that researchers
can't always get ideal populations so they study the populations they
have. Unfortunately the study is behind a paywall, so you can't see how it was
designed, although you would think the Gruniad would not report on
something that was obviously flawed.
Here's another study: "This study used a pretest/posttest design and included a control group
to examine the impact of harassment training on intended
responses to harassment. The sample consisted of
282 full-time professionals. At time 2, trainees expressed lower
intentions
to confront the perpetrator than did control-group
participants." [1]
This one, "Sexual harassment at work, a decade (plus) of progress" has been cited widely. [2]